International Journal of Advanced Mechatronic Systems (5 papers in press)

Regular Issues

• Learning to land: autonomous quadcopter recovery from rotor loss using adaptive thrust vectoring  
  by Zairil Zaludin 
  Abstract: Quadcopter drones depend on four rotors to manage altitude and orientation. However, the failure of a single rotor compromises the drone's ability to remain airborne and to land safely. This article introduces a solution enhancing attitude control in the event of a complete rotor failure by reducing roll, pitch, and yaw deviations during landing. This was achieved by actively tilting and panning the three remaining operational rotors. The controller for the pan and tilt mechanism was developed using the reinforcement learning approach. The solution was reached after the agent accumulated reward points over 4,000 training episodes when the feedforward thrust setting for the rotors was set to '48'. The uncontrollable attitude during flight was mitigated. The experiment was extended to explore the impact of changing the feedforward thrust setting to '60'. With the additional thrust, the unbalanced drone demonstrated better attitude responses during touchdown.
  Keywords: autonomous quadcopter recovery; rotor loss; adaptive thrust vectoring; single rotor failure; rotor pan; rotor tilt; reinforcement learning; deep deterministic policy gradient; DDPG.
  DOI: 10.1504/IJAMECHS.2026.10074075
   
  
• Optimised attention-based CNN for crowd density estimation with wiener filtering  
  by Jyoti Ambadas Kendule, Kailash J. Karande 
  Abstract: Crowd density estimation, the process of quantifying individuals in a given space, is crucial for urban planning, public safety, and event management. Accurate estimation supports effective decision-making, resource allocation, and crowd control. Traditional methods often rely on manual counting or simple techniques that fail to address the complexities of crowded scenes. To overcome these limitations, this paper introduces a novel deep learning-based crowd density estimation (DL-CDE) framework. The framework begins by converting video footage of the crowd into individual frames, enabling detailed analysis. These frames undergo pre-processing with a root mean square assisted wiener filtering (RAWF) technique, which enhances clarity and reduces noise distortion. The processed frames are then fed into a modified attention mechanism-based convolutional neural network (MA-CNN), which performs both object detection and crowd density estimation. This approach provides more accurate and reliable crowd density predictions, improving the effectiveness of crowd management and safety measures in various settings.
  Keywords: crowd density estimation; deep learning; DL; root mean square assisted wiener filtering; RAWF-based pre-processing; mechanism-based convolutional neural network; MA-CNN; attention mechanism.
  
  
• Harmonic compensation and closed-loop control for single-phase meter calibration power supply  
  by Fuzhuan Wu, Manman Li, Shengjun Wen, Yongsheng Zhu 
  Abstract: This paper proposes a three-closed-loop control strategy based on harmonic compensation to address the requirements of voltage and current sources with wide output ranges and low harmonic content in metre calibration power supply. Firstly, a root mean square (RMS) value controller is designed in outer loop to control the output voltage and current. Then, a harmonic compensator in the instantaneous value middle loop is combined with a proportional-integral controller in the inductor current inner loop to suppress disturbances between the power supply and the load, as well as reduce the increase in harmonic content caused by factors such as inductor saturation, dead time settings, and transformer magnetic saturation. Also, a phase closed-loop regulator based on over-frequency sampling is investigated to ensure precise regulation of the phase in voltage and current sources, which optimises the selection of the zero-crossing point by increasing the A/D sampling frequency and applying closed-loop regulation to the phase difference between the voltage and current sources and a given value. Finally, the effectiveness of the proposed control strategy is verified through MATLAB/Simulink simulation and a prototype system with DSP28335.
  Keywords: harmonic compensation; three closed-loop control; over-frequency sampling; closed-loop phase control; root mean square; RMS.
  DOI: 10.1504/IJAMECHS.2026.10074470
   
  
• A novel blockchain approach to identify child malnutrition using residual pyramid forward fractional network  
  by Prateeksha Chouksey, Prasadu Peddi, Sandeep Kadam 
  Abstract: Malnutrition occurs due to a lack of nutrients, which is often curable with early detection. However, detecting malnutrition at an early stage is challenging and ineffective in various techniques. Overfitting also affects the previous method's performance. Therefore, here the residual pyramid forward fractional network (RPFF-Net) is developed for child malnutrition. Initially, the child's health is recorded in the blocks of the blockchain. Then, the recorded data is saved in the cloud server, which holds the tracking architecture. Next, the data is normalised by Z-score normalisation. After that, the features are selected by employing an ensemble-based model with mutual information, information gain (IG), and recursive feature elimination (RFE). After that, nutrition status is tracked by the RPFF-Net approach, which is created by fusing PyramidNet, deep residual network (DRN), and fractional calculus (FC). The RPFF-Net attained a true positive rate (TPR), accuracy, and true negative rate (TNR) of 91.87%, 90.59%, and 90.98%.
  Keywords: malnutrition; Z-score normalisation; deep residual network; DNR; pyramid network; fractional calculus; FC.
  DOI: 10.1504/IJAMECHS.2026.10072878
   
  
• Hybridisation of nonlinear autoregressive model with deep long short-term memory network for crop damage detection using time series data  
  by Saravanan Radhakrishnan, V. Vijayarajan 
  Abstract: Climate change and population growth have intensified crop damage in recent years. This paper introduced nonlinear autoregressive model with exogenous inputs fused with deep long-short term memory (NARX-DLSTM) for detecting crop damage at an early stage. The input time series data undergoes preprocessing, followed by the extraction of technical indicators like relative strength index (RSI), double exponential moving average (DEMA), weighted moving average (WMA), simple moving average (SMA), Welles Wilder's smoothing average (WWS), moving average convergence divergence (MACD), linear regression forecast (LRF) and lowest low value (LL). Then, feature selection is performed using weighted Euclidean distance (WED), and data augmentation is applied through synthetic minority over-sampling technique (SMOTE). Finally, NARX-DLSTM is performed which is the combination of DLSTM and NARX recurrent neural networks, achieves mean squared error (MSE), root mean square error (RMSE), mean absolute percentage error (MAPE), and relative absolute error (RAE) of 0.200, 0.447, 0.165, and 0.114.
  Keywords: weighted Euclidean distance; WED; crop damage detection; deep long-short term memory; DLSTM; synthetic minority over-sampling technique; SMOTE; recurrent neural networks; RNNs.
  DOI: 10.1504/IJAMECHS.2026.10073117